week ending

PRL 118, 244801 (2017) PHYSICAL REVIEW LETTERS 16 JUNE 2017

Ion Motion Induced Emittance Growth of Matched Electron Beams in Plasma Wakefields
Weiming An,1,2,* Wei Lu,3,4 Chengkun Huang,5 Xinlu Xu,1,2 Mark J. Hogan,6 Chan Joshi,1 and Warren B. Mori1,2
1
Department of Electrical Engineering, University of California Los Angeles, Los Angeles, California 90095, USA
2
Department of Physics and Astronomy, University of California Los Angeles, Los Angeles, California 90095, USA
3
Department of Engineering Physics, Tsinghua University, Beijing 100084, China
4
IFSA Collaborative Innovation Center, Shanghai Jiao Tong University, Shanghai 200240, China
5
Los Alamos National Laboratory, Los Alamos, New Mexico 87545, USA
6
SLAC National Accelerator Laboratory, Menlo Park, California 94025, USA
(Received 4 March 2016; revised manuscript received 27 March 2017; published 14 June 2017)
Plasma-based acceleration is being considered as the basis for building a future linear collider. Nonlinear
plasma wakefields have ideal properties for accelerating and focusing electron beams. Preservation of the
emittance of nano-Coulomb beams with nanometer scale matched spot sizes in these wakefields remains a
critical issue due to ion motion caused by their large space charge forces. We use fully resolved quasistatic
particle-in-cell simulations of electron beams in hydrogen and lithium plasmas, including when the
accelerated beam has different emittances in the two transverse planes. The projected emittance initially
grows and rapidly saturates with a maximum emittance growth of less than 80% in hydrogen and 20%
in lithium. The use of overfocused beams is found to dramatically reduce the emittance growth.
The underlying physics that leads to the lower than expected emittance growth is elucidated.

DOI: 10.1103/PhysRevLett.118.244801

There has been tremendous recent progress on plasma-
based acceleration (PBA) where charged particles are accel-
erated by plasma wakefields [1–8]. This has led to the
consideration of designing a future linear collider (LC) using
particle beams or lasers to drive the wakefields [9–11]. With
respect to particle beam-driven plasma wakefield acceler-
ation (PWFA), this progress includes demonstrating sus-
tained high gradient acceleration (∼50 GeV=m) over 1 m [5],
efficient transfer of energy from the wake to the trailing beam
[6] and high gradient positron acceleration in self-loaded
wakes [8]. For future LC designs, it is also necessary that
the luminosity, L ¼ ðfN 2 =4πσ x σ y Þ be large, where N is the
number of particles in each bunch, f is the frequency of the
collisions, and σ x;y is the spot size of the bunch at
the interaction point. In order to focus a bunch to the spot
size needed to achieve luminosities ∼1034 cm−2 s−1 , the
bunch should have a normalized emittance, ϵN , of ∼100 nm
and ∼1 nC of charge.
Much research in PBA has been on nonlinear wakefields
[12–14] because they have ideal properties for accelerating
and focusing electron beams with nC of charge. They are
described by complete electron blowout. The blown out
electrons form a narrow sheath with the shape of a “bubble”
that surrounds the plasma ions. If the wake is axisymmetric
and plasma ions do not move, the focusing field can be
shown to be Ff ¼ me ω2p r=2 [12–14], which is proportional
to r and independent of ξ ¼ ct − z, where ωp is the plasma
frequency. Unless specified, henceforth we normalize the
length to c=ωp ≡ k−1 p , the time to 1=ωp , the mass to the
electron mass, me , and the charge to e. For central forces
that are linear in r, the forces in the x̂ and ŷ directions are
also linear in x and y, respectively. The matched spot size is
σ r ¼ ð2ϵ2N =γÞ1=4 [15], where γ is the relativistic Lorentz
factor of the beam, which corresponds to 100 nm size
beams for the emittances required to reach the luminosities
of future LC parameters.
In Ref. [16] it was shown that, for tightly focused nC class
beams, the Coulomb field can pull the ions inward during the
transit time of the beam. Essentially, when nb =n0 ≫ mi =me
(where n0 is the plasma density), the ion density is com-
pressed, which perturbs the focusing force, Ff ¼ r=2 þ F~ f ,
potentially causing catastrophic emittance growth. In
Ref. [17] it was proposed to adiabatically match the beam
to the perturbed focusing field by using plasma sections with
decreasing plasma ion masses. However, there has been little
work on quantifying the emittance growth from the ion
motion itself because the simulations need to be self-
consistent and resolve the small spot sizes of the beam while
including the much larger size of the bubble.
In this Letter, we investigate the ion motion induced
emittance growth through the use of highly resolved
QuickPIC [18,19] simulations and single particle calculations
in parametrized focusing fields. QuickPIC is a three-
dimensional quasistatic particle-in-cell code, which is used
to efficiently model short-pulse laser and particle beam
plasma interactions [20]. The results show that, even
though the ion density compression within the e− bunch
is large, the emittance growth is modest because the ion
collapse is only strong in the core of the beam, the single
particle orbits are highly anharmonic, and the ion collapse
varies along the bunch length. We also find that over-
focusing the accelerated bunch (using a spot size less than
the matched spot size) can also dramatically reduce the
emittance growth.

0031-9007=17=118(24)=244801(5) 244801-1 © 2017 American Physical Society


PRL 118, 244801 (2017) PHYSICAL REVIEW LETTERS
(a) (b)
FIG. 1. Emittance evolution of a beam slice in a fixed nonlinear
focusing field. (a) Initial beam density profile in arbitrary units
(the gray dashed line), the focusing field (FF, the red line with
A ¼ 200 and the red dashed line with A ¼ 0), and the particle
trajectories (Traj.) in the focusing field of A ¼ 200 (the blue solid
line) and A ¼ 0 (the blue dashed line). (b) The
pffiffiffislice
ffi beam
emittance evolution and the emittance growth vs A.
We start by studying how a given longitudinal slice
of the trailing beam evolves in prescribed nonlinear
focusing fields due to an ion density peak on axis of the
form nion =n0 ¼ 1 þ A expð−r2 =2σ 2 Þ, for which the
corresponding focusing force is Ff ¼ fr=2 þ Aσ 2 ½1−
expð−r2 =2σ 2 Þ=rgr̂. For nonlinear forces, the motions in
the x̂ and ŷ directions are coupled. For simplicity, we begin
by considering motion only in the x̂ direction (we also
neglect to lowest order the changes to γ); i.e., we consider
particles with y ¼ py ¼ 0. They feel the maximum jF⃗ f · x̂j
compared to those at the same x but different y and py and
hence have the largest perturbation on their phase space
distribution. We rely on the numerical simulations to
include the coupling between the two planes and changes
to γ. The insight gained from the 1D model is born out in
the simulations.
We use the particle push in QuickPIC to simulate the
evolution of a slice of a beam in a plasma with
n0 ¼ 1.0 × 1017 cm−3 . The beam slice has an initial
normalized emittance ϵNx0 ¼ 0.1 μm, where ϵNx ≡
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
hx2 ihp2x i − hxpx i2 (px is normalized to me c) and hi
represents an average over particles. The initial spot size
is σ x0 ¼ 0.104 μm and the initial energy is 25 GeV, so
the beam is initially matched to the x=2 focusing field. We
choose the width of the ion density peak as σ ¼ 0.104 μm,
which was set equal to σ x0 . The energy of the beam
particles is kept constant. In Fig. 1(a), we plot the phase
space trajectories, x-px , for a particle initially located at
x ¼ σ x0 , px ¼ 0 and under the focusing field with A ¼ 0
and A ¼ 200, respectively.
When the focusing field is perfectly linear (i.e., A ¼ 0),
the maximum momentum px0 of the particle is equal to
0.964, which satisfies σ x0 px0 ¼ ϵNx0 because the slice
beam is matched. When there is ion collapse, px0 will
be larger because jFf j is larger. As shown in Fig. 1(a), px0
pffiffiffiffi
increases approximately by a factor of A ≈ 14.2 since the
slope of the focusing field inside the beam center is around
244801-2
week ending
16 JUNE 2017
A times larger. Since Ff is nonlinear, the saturated
pffiffiffiffi
emittance growth will also scale with A with a propor-
tionality constant less than unity that depends on σ=σ x0 .
In Fig. 1(b), we plot the emittance growth for different A’s
but with the same σ=σ px0ffiffiffiffi, and the saturated emittance is seen
to beproughly
ffiffiffiffi 0.425 A. This scaling of emittance growth
with A, i.e., with the square root of the peak ion density,
would be catastrophic.
However, this is not what is observed in properly
resolved self-consistent simulations. We consider the fol-
lowing beam loading scenario where a drive beam with
3.0 × 1010 electrons transfers energy to a trailing beam with
1.0 × 1010 electrons [9,21]. The drive and trailing beams
have ϵNd ¼ 1 mm and ϵNt ¼ 0.1 μm, respectively. The
emittance of the trailing beam is chosen to achieve
the necessary luminosity of a LC [10]. Both the drive
and the trailing beams have an initial energy of 25 GeV
2 2 2 2
and a bi-Gaussian density profile, ∼e−r =ð2σ r Þ e−z =ð2σz Þ . The
17 −3
plasma density is n0 ¼ 1.0 × 10 cm . The matched rms
spot sizes and pulse lengths of the drive and trailing beams
are σ rd ¼ 10.37 μm and σ rt ¼ 0.1 μm, and σ zd ¼ 30 μm
and σ zt ¼ 10 μm, respectively. The ratio ðnb =n0 Þ of the
trailing beam exceeds mi =me for hydrogen, so ion motion
is expected to be an issue. Physically, ion motion occurs
within the beam when the ion plasma
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi frequency for the
beam density, Ωb ¼ 4πnb e2 =mi , times the beam duration
exceeds unity, i.e., Ωb σ z =c ≫ 1. This leads to the previous
condition if σ z ≈ c=ωp . The two beams are separated by
115 μm so that the trailing beam can flatten the longitudinal
electric field Ez of the wake in order to maintain a small
energy spread [22], and the flattened Ez ≈ −1.0.
We carried out a QuickPIC simulation using a box
with 8192 × 8192 × 1024 cells, and the resolution is
48.83 × 48.83 × 305.66 nm. In Fig. 2(a), 2D isosurface
plots of the plasma and beam densities are shown for a
slice along y ¼ 0. The bubble radius is seen to be
∼3.5c=ωp ≈ 54 μm, which is more than 500 times larger
than σ rt . This simulation also showed significant ion
collapse, as well as rapid emittance growth. However,
there were only two cells across σ rt . We therefore carried
out simulations that only modeled a small volume [the
dimensions are shown as a red box in Fig. 2(a)], but with a
much higher resolution of 6.10 × 6.10 × 115.23 nm.
The plasma was initialized as an ion column surrounded
by a neutralizing and stationary electron sheath. We assume
that the drive beam does not contribute to the ion motion.
The ion collapse and the focusing field on the trailing beam
were nearly identical to that in the two-beam simulation
with the same resolution [20].
In Fig. 2(b), we show the ion (Hþ ) density behind the
trailing beam (centered at ξ ¼ 0) in the y ¼ 0 slice.
Following an initial spike, the peak ion density increases
through the beam (along ξ), albeit with an oscillation to
a maximum value around 200 times larger than n0 .

PRL 118, 244801 (2017)
(a) 17 -3
-10
Plasma electron density (10 cm )
-8 -6 -4 -2 0 0
6 0
4
2 150
X (c / ω p)
0
-2
-4
-6
-12 -10 -8 -6 -4 -2 0 -1.0
ξ (c / ω p )
(c)
FIG. 2. PWFA with ion motion. (a) Nonlinear wake in H plasma
and the drive and trailing beam densities (ξ ¼ 0 is the center of
the trailing beam). (b) Plasma ion density in x-ξ plane (ξ ¼ 0 is
the center of the trailing beam). (c) Ff transverse lineouts at
different ξ’s and the initial beam density profile. (d) The trailing
beam’s projected and slice emittance evolutions. The plasma skin
depth is k−1
p ¼ 16.83 μm in these plots.
The corresponding Ff is significantly perturbed around the
axis, as seen in Fig. 2(c), where F⃗ f · x̂ in the y ¼ 0 plane is
plotted for several values of ξ. The slope of F⃗ f · x̂ is nearly
the same in each slice, but the maximum value and width
grows through the beam. The initial transverse density
profile of the beam is shown as a reference. From Fig. 2(c),
we can also estimate the Rperturbation of Ez using Panofsky-
Wenzel theorem ΔEz ¼ dr∂Ff =∂ξ≈ΔrΔFf =2Δξ, which
is on the order of 0.002 for Δr ¼ 0.1, ΔFf ¼ 0.02, and
Δξ ¼ 0.5. The ΔEz is negligible compared to −1.0, which
is that felt by the trailing beam. This is consistent with the
lower resolution simulations where the bubble excitation is
also modeled. The basic reason for such a small perturba-
tion on Ez is that, for each slice, the total charge contained
in the ion density peak is very small, and it changes slowly
along ξ.
In Fig. 2(d), we plot the emittance growth for slices at the
same values of ξ, as well as the projected emittance. The
emittance is seen to rapidly grow and then saturate for each
slice. The projected emittance (and the slice in the middle
of the beam) grows by less than a factor of 1.8, and slices in
the rear of the beam grow only by a factor of 2.1. This
emittance growth is much less than the anticipated growth
[16,17] and that seen in Fig. 1(b) for A ¼ 200 and σ ¼ σ x0 .
The fundamental reason for the significantly smaller than
expected emittance growth is that the ion compression is
much narrower than the initial beam spot size. This can be
week ending
PHYSICAL REVIEW LETTERS 16 JUNE 2017
(b)
50
Plasma ion density
100 150 200 seen in Fig. 3(a), where a lineout of the ion density vs x (for
y ¼ 0) is shown in the middle of the beam; i.e., ξ ¼ 0 is
Ion density (10 177 cm )
-3
-2
200 shown from the above simulation. For comparison, the
Beam density (10 cm )
-3
initial trailing beam profile (the dashed gray line) and the
17
-4
100 prescribed form for the ion density (the dashed red line) for
-6 50
0
A ¼ 135.9 and σ ¼ 0.1σ x0 are also shown. The narrower
1.0
-8
0.5 ion compression leads to a smaller value of px0 and to an
0.0 0.02
anharmonic motion, such that the time average over a
)
0.01
p
pffiffiffi
/ω
-10 -0.5 0.00 )
(c
-0.01 (c / ω p
particle’s orbit is less than px0 = 2. We note that the ion
ξ
-0.02 X
(d) collapse develops a pedestal outside the core as one moves
through the bunch, although the width of the core remains
unchanged. This effectively increases σ for the later slices.
To quantify the emittance growth, we first note that
just as the emittance quickly reaches a steady state [as seen
in Fig. 2(d)], so too does the beam phase space and the
ion density. In the steady state (where the spot size does
not change), hxpx if ¼ 0, so the final emittance of the beam
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
is ϵNxf ¼ hx2 if hp2x if . In addition, in the steady state,
hx2 if ¼ fhx2 igt;Δt ¼ hfx2 gt;Δt i, where fgt;Δt represents
the time average of a quantity at time t during a duration
Δt. We can choose a Δt ¼ T that is much larger
than every particle’s oscillation R R x0 dx τ. Therefore,
period R x0
fx2 gt;T ¼ fx2 gt;τ ≡ X2ave ¼ ð 0x0 dx x2
= 0 vx Þ ¼ ð 0 dxx =
2
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi R x0 vx
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
½ψðx; ξÞ − ψðx0 ; ξÞÞ = ð 0 dx = ½ψðx; ξÞ − ψðx0 ; ξÞÞ,
where x0 is the maximum value of x, vx ¼ px =γ ¼
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2½ψðx; ξÞ − ψðx0 ; ξÞ=γ , and ψ is the wake potential
[Ff ðxÞ ¼ −∂ψ=∂x]. For highly relativistic beams, there is
no phase slippage, so each slice evolves independently with
a different phase space distribution. We henceforth assume
that γ does not change, so it can be brought out of the
integrals. In reality, γ changes adiabatically, and including
this in the numerical work does not alter the results.
Following analogous reasoning leads to fp2x gt;T ¼P2ave ¼
R pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi R pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
γð 0x0dx ½ψðx;ξÞ−ψðx0 ;ξÞÞ=ð x00 dx= ½ψðx;ξÞ−ψðx0 ;ξÞÞ.
In Fig. 3(b), we plot px and x vs s ¼ z ≈ ct for an
electron starting at rest at x0 ¼ σ x0 in a focusing force with
A ¼ 200 and σ ¼ 0.1σ x0 or σ ¼ σ x0. The s axis is normal-
ized to the period of the oscillation for each case, while px
and x are normalized to their maximum values px0 and x0 .
It is clearly seen that the xðsÞ motion is essentially
harmonic for both cases, while the px ðsÞ motion is very
different for the σ ¼ 0.1σ x0 (narrow ion collapse) case; i.e.,
it is anharmonic. pffiffiffi Because the xðsÞ motion is harmonic,
Xave =x0 ≈ 1= 2 for bothpcases, ffiffiffi while, by inspection of
Fig. 3(b), Pave =px0 ≪ 1= 2 when σ ≪ σ x0 . To quantify
this, in Fig. 3(c), we plot how Pave, px0 , and Xave depend on
x0 and σ for A ¼ 200. This clearly shows that Pave is much
smaller than px0 and that px0 is much smaller when the ion
collapse is narrower.
We now use Pave , Xave , and the initial beam distribution
function, f 0 ðx; pÞ, to calculate hx2 if and hp2 if . As men-
tioned before, hx2 if ¼ hX2ave i. To calculate hX2ave i, we sort
244801-3

PRL 118, 244801 (2017) PHYSICAL REVIEW LETTERS
(a) (b)
(c) (d)
FIG. 3. (a) Plasma ion density lineout at ξ ¼ 0 (the blue dashed
line) compared with the initial trailing beam distribution (the gray
dashed line) and a Gaussian fit for the core of the ion density (the
red dashed line). (b) px and x vs s in different focusing field (the
solid line for σ=σ x0 ¼ 1.0 and the dotted line for σ=σ x0 ¼ 0.1) for
the particle initially located at x ¼ σ x0 . (c) X ave ðx0 Þ, Pave ðx0 Þ,
px0 ðx0 Þ, and N x0 (in an arbitrary unit) in the focusing field with
A ¼ 200 and different σ’s. (d) Slice beam emittance evolution in
the focusing field with A ¼ 200 and different σ’s and final
emittance growth from the simulation and theory.
all the electrons by their orbits in phase space. All electrons
with a maximum x value between x0 and x0 þ dx0 ,
where dx0 ≪ x0 , will have 2 plane with the larger emittance, ϵNx , grows by only 10%. In
R ∞ the same2Xave ðx0 Þ. Therefore,
2 2
hx if ¼ hXave i ¼ ð1=NÞ 0 dx0 N x0 Xave ðx0 Þ, where N is
the total particle number and N x0 dx0 is the number of
particles with orbits that have a maximum x between
x0 and x0 þ dx0 . N x0 will not change in time and, according
to its definition, it canR be calculated using f 0 ðx; pÞ. It
follows that N x0 ¼ 4 0x0 dxf 0 ðx; px Þð∂px =∂x0 Þ, where
pffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
∂px =∂x0 ¼ γ Ff ðx0 Þ= 2½ψðxÞ − ψðx0 Þ. We can use
f 0 since the ion collapse is in a quasi steady state
even during the duration of the initial emittance
growth. Following R the same reasoning, we have
hp2x if ¼ ð1=NÞ 0∞ dx0 N x0 P2ave ðx0 Þ. In Fig. 3(c), we also
plot N x0 for a case where there is no ion collapse (A ¼ 0),
and for A ¼ 200 with different σ’s. This plot shows that
when there is ion collapse the ensemble average is
weighted towards particles with smaller x0 ’s.
The final emittance (assuming 1D-like motion) can be
predicted for each slice using only ψ and f 0 . In Fig. 3(d),
we show the predicted final emittance growth (compared to
the simulated results) for slices with initially Gaussian
beams with matched spot sizes for the parameters described
earlier. The curves correspond to different widths for the
ion collapse with A ¼ 200. The curves show significantly
smaller emittancepgrowth
ffiffiffiffi than what would be naively
predicted from a A scaling.
244801-4
week ending
16 JUNE 2017
(a) (b)
FIG. 4. Emittance growth of (a) an asymmetric beam in
hydrogen and lithium and (b) a symmetric beam with an initially
unmatched beam spot size in hydrogen (compared to the initially
matched case).
In LC designs, asymmetric emittances are used to
minimize the beamstrahlung [23] that occurs during dis-
ruption [24] at the final focus. In such a case, the ion
collapse also becomes asymmetric, and developing a model
for the emittance growth is much more complicated.
Therefore, we rely on highly resolved QuickPIC simulations.
In Fig. 4, we plot the emittance growth for a case where
the initial emittances in the two transverse planes are
pffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ϵNx ¼ 2.0 μm, ϵNy ¼ 0.005 μm ( ϵNx ϵNy ¼ 0.1μm is the
same as before). We still match the initial beam spot sizes
to the r=2 focusing field, so σ rtx ¼ 463.9 nm and
σ rty ¼ 23.2 nm. The solid red and blue lines show the
evolution of the trailing beam’s projected emittances for a
hydrogen plasma.
It can be seen that the projected emittance growth in the
the other plane, ϵNy grows by 120%. Therefore, the growth
pffiffiffiffiffiffiffiffiffiffiffiffiffiffi
of ϵNx ϵNy is only 55.6%, which is even smaller than the
symmetric case. This appears to be mostly due to the fact
that the peak ion density is smaller. To further mitigate the
emittance growth, one could use a heavier ion, such as Liþ .
In Fig. 4(a), we show the emittance evolution of the same
beam in a lithium plasma as the red and blue dotted lines.
The emittance growth in the two planes is now less than 1%
and 40%, respectively, which are both reduced compared to
pffiffiffiffiffiffiffiffiffiffiffiffiffiffi
the hydrogen plasma. The growth of ϵNx ϵNy for Li is only
18.9%. An issue for a Li plasma that requires further
investigation is that the tightly focused trailing beam may
further ionize the Liþ ion if the transverse electric field
exceeds ∼400 GV=m, which would cause additional mod-
ifications to the focusing force.
Another way to mitigate emittance growth is to use an
initial beam spot size less than the matched spot size
(overfocused) for fixed ions, but closer to the steady state
value when there is ion collapse. Figure 4(b) shows the
projected emittance growth for a symmetric beam with an
initially unmatched spot size. The beam has the same initial
emittance used to generate Fig. 2, but it has a smaller spot
size, σ rt ¼ 50.0 nm. The projected emittance growth is
substantially less, but it now appears to oscillate. The
amplitude of the oscillations decreases over time, indicating

PRL 118, 244801 (2017) PHYSICAL REVIEW LETTERS
that a steady state is being approached with a final
emittance growth of only ∼25%. Finally, we note that in
a PWFA LC, a series of stages will be used. These results
indicate that the emittance growth might occur only in the
first stage and that, once the steady state is formed, the
emittance will not grow in later stages if the beam can be
properly transported into and out of the additional stages.
This work was supported by the U.S. Department of
Energy under Grants No. DE-SC0010064, No. DE-
SC0014260, and No. DE-SC0008316, and NSF Grants
No. ACI-1339893 and No. PHY-1500630, and NSFC
Grants No. 11425521 and No. 11535006. The simulations
were carried out on the Blue Waters under NSF Grant
No. ACI-1614949 and the UCLA Hoffman 2 and Dawson 2
clusters.
* J. Comput. Phys. 217, 658 (2006).
anweiming@ucla.edu
[1] C. G. R. Geddes, Cs. Toth, J. van Tilborg, E. Esarey, C. B.
Schroeder, D. Bruhwiler, C. Nieter, J. Cary, and W. P.
Leemans, Nature (London) 431, 538 (2004).
[2] J. Faure, Y. Glinec, A. Pukhov, S. Kiselev, S. Gordienko, E.
Lefebvre, J.-P. Rousseau, F. Burgy, and V. Malka, Nature
(London) 431, 541 (2004).
[3] S. P. D. Mangles et al., Nature (London) 431, 535
(2004).
[4] W. P. Leemans et al., Nat. Phys. 2, 696 (2006).
[5] I. Blumenfeld et al., Nature (London) 445, 741 (2007).
[6] M. Litos et al., Nature (London) 515, 92 (2014).
[7] W. P. Leemans et al., Phys. Rev. Lett. 113, 245002
(2014).
[8] S. Corde et al., Nature (London) 524, 442 (2015).
244801-5
week ending
16 JUNE 2017
[9] A. Seryi et al., in Proceedings of the 2009 PAC Conference,
Vancouver, Canada, 2009, edited by M. Comyn, S.
Koscielniak, V. R. W. Schaa, and P. W. Schmor (AIP,
New York, 2009).
[10] E. Adli et al., arXiv:1308.1145.
[11] W. Leemans and E. Esarey, Phys. Today 62, 44 (2009).
[12] J. B. Rosenzweig, B. Breizman, T. Katsouleas, and J. J. Su,
Phys. Rev. A 44, R6189(R) (1991).
[13] W. Lu, C. Huang, M. Zhou, W. B. Mori, and T. Katsouleas,
Phys. Rev. Lett. 96, 165002 (2006).
[14] W. Lu, C. Huang, M. Zhou, M. Tzoufras, F. S. Tsung,
W. B. Mori, and T. Katsouleas, Phys. Plasmas 13, 056709
(2006).
[15] C. Joshi et al., Phys. Plasmas 9, 1845 (2002).
[16] J. B. Rosenzweig, A. M. Cook, A. Scott, M. C. Thompson,
and R. B. Yoder, Phys. Rev. Lett. 95, 195002 (2005).
[17] R. Gholizadeh, T. Katsouleas, P. Muggli, C. Huang, and
W. Mori, Phys. Rev. Lett. 104, 155001 (2010).
[18] C. Huang, V. K. Decyk, C. Ren, M. Zhou, W. Lu, W. B.
Mori, J. H. Cooley, T. M. Antonsen, and T. Katsouleas,
[19] W. An, V. K. Decyk, W. B. Mori, and T. M. Antonsen,
J. Comput. Phys. 250, 165 (2013).
[20] See Supplemental Material at http://link.aps.org/
supplemental/10.1103/PhysRevLett.118.244801 for
QuickPIC and its simulation.
[21] C. Huang et al., in Proceedings of the 2009 PAC
Conference, Vancouver, Canada, 2009, edited by M.
Comyn, S. Koscielniak, V. R. W. Schaa, and P. W. Schmor
(AIP, New York, 2009).
[22] M. Tzoufras, W. Lu, F. S. Tsung, C. Huang, W. B. Mori,
T. Katsouleas, J. Vieira, R. A. Fonseca, and L. O. Silva,
Phys. Rev. Lett. 101, 145002 (2008).
[23] M. Bell and J. S. Bell, Part. Accel. 24, 1 (1988).
[24] P. Chen and K. Yokoya, Phys. Rev. D 38, 987 (1988).